]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
md: fix create on open mddev lifetime regression
authorYu Kuai <yukuai3@huawei.com>
Wed, 30 Jul 2025 07:33:21 +0000 (15:33 +0800)
committerYu Kuai <yukuai3@huawei.com>
Wed, 30 Jul 2025 17:21:43 +0000 (01:21 +0800)
Commit 9e59d609763f ("md: call del_gendisk in control path") moves
setting MD_DELETED from __mddev_put() to do_md_stop(), however, for the
case create on open, mddev can be freed without do_md_stop():

1) open

md_probe
 md_alloc_and_put
  md_alloc
   mddev_alloc
   atomic_set(&mddev->active, 1);
   mddev->hold_active = UNTIL_IOCTL
  mddev_put
   atomic_dec_and_test(&mddev->active)
    if (mddev->hold_active)
    -> active is 0, hold_active is set
md_open
 mddev_get
  atomic_inc(&mddev->active);

2) ioctl that is not STOP_ARRAY, for example, GET_ARRAY_INFO:

md_ioctl
 mddev->hold_active = 0

3) close

md_release
 mddev_put(mddev);
  atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)
  __mddev_put
  -> hold_active is cleared, mddev will be freed
  queue_work(md_misc_wq, &mddev->del_work)

Now that MD_DELETED is not set, before mddev is freed by
mddev_delayed_delete(), md_open can still succeed and break mddev
lifetime, causing mddev->kobj refcount underflow or mddev uaf
problem.

Fix this problem by setting MD_DELETED before queuing del_work.

Reported-by: syzbot+9921e319bd6168140b40@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/68894408.a00a0220.26d0e1.0012.GAE@google.com/
Reported-by: syzbot+fa3a12519f0d3fd4ec16@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/68894408.a00a0220.26d0e1.0013.GAE@google.com/
Fixes: 9e59d609763f ("md: call del_gendisk in control path")
Link: https://lore.kernel.org/linux-raid/20250730073321.2583158-1-yukuai1@huaweicloud.com
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Xiao Ni <xni@redhat.com>
drivers/md/md.c

index 046fe85c76fe18de7370f8660b56048b5a880394..2716d5c59517ab3bcc8f500b26ba755791017957 100644 (file)
@@ -636,6 +636,12 @@ static void __mddev_put(struct mddev *mddev)
            mddev->ctime || mddev->hold_active)
                return;
 
+       /*
+        * If array is freed by stopping array, MD_DELETED is set by
+        * do_md_stop(), MD_DELETED is still set here in case mddev is freed
+        * directly by closing a mddev that is created by create_on_open.
+        */
+       set_bit(MD_DELETED, &mddev->flags);
        /*
         * Call queue_work inside the spinlock so that flush_workqueue() after
         * mddev_find will succeed in waiting for the work to be done.